* It is a trie. The max depth of the trie is the length of the
longest prefix. The lookup is O(lookuped_prefix), but as the prefix
length is limited, the lookup time is constant.
* Each node can hold the entire prefix, has place for the rewrite
rule with up to three digits.
* A trie with 20k entries will take about 3MB ram.
* Filling the trie 100 times takes ~800ms on my i7 laptop
* 10.000.000 lookups take 315ms.. (for the same prefix).
* 93/99 lines are tested, 6/6 functions are tested, 49 of 54 branches
are tested. Only memory allocation failures are not covered
* A late addition is to handle the '+' sign and to increase the number
of chars in the rewrite prefix. The timing/line coverage has not
been updated after this change.
The jenkins build node has Python 2.5.X installed and the
assertGreater method is not available. Use assert_ until
we can use newer versions of Python.
Disable the periodic LU using "no periodic location update" VTY
command. In that case set the expire_lu to 0 which will then be
translated to a NULL in the database layer. This leads to a bit of
copy and paste in the db_sync_subscriber method but I don't see
how we could easily use 'datetime(%i, 'unixepoch')' and 'NULL'
at the same time.
Change the query to find expired queries to check for NOT NULL
and the time being in the past. This means if there are still
old subscribers in the database they might not be expired. One
would need to execute a query like "UPATE Subscriber SET expire_lu
= 0 WHERE expire_lu is null". The same applies when disabling the
periodic LU. One would need to update the database by hand.
Manual tests executed/passed:
1.) periodic LU enabled:
* use gst LUTest.st to do a LU
* UPDATE Subscriber SET expire_lu=datetime('now');
* observe the subscriber being expired (it was)
2.) periodic LU disabled:
* use gst LUTest.st to do a LU
* verify that the expire_lu is NULL in the database
We were expiring subscribers during active calls. This is because
the T3212 is stopped under certain conditions but we didn't stop
that timer at all.
Remember if T3212 timer was stopped due something done by NITB and
update the expiration time at the end of the radio connection, as
the phone should restart it when returning to MM Idle.
It is a bit difficult to decide when we should set the flag. E.g.
in a CM Service Request we don't know if we accept the service and
during a LU we already send MM messages before we accept or reject
the subscriber.
The easiest is to set the flag when receiving a paging response
on known subscribers and at the end of the authentication process.
Do not expire a subscriber that has an active connection that is
marked with the flag, e.g. we would still expire a subscriber that
is being paged.
Manual tests executed/passed:
* gst LUTest.st verified that a expiration date was set
* gst SMSTest.st (doing another LU but forcing a timeout on the
SMS sending). Verified that the expire_lu was updated.
Make sure to not ever have issues with this code again, move the
utility code to a new file and create a basic testcase. The method
currently has 100% line and branch coverage. My initial patched
missed the smpp_utils.c file and I re-did the copying (and verifying
the branch coverage)
Begin with the NAT code. It is not clear yet if we will have one
file with all the tests or will have a sub directory with *.py files.
In the long run the base class will move to the osmo-python-tests
module.
The test scripts warn about missing documentation, untested configs,
check common errors, and stub out testing individual VTY commands.
The scripts have been moved to the another osmocom repository,
python/osmo-python-tests
The features were requested by zecke.
Find the Cell Identifier from the Complete Layer3 Information and
store it for future reference. We could begin to verify that the
LAC/CI used really belongs to the BSC.
The name sccp_connection is used in the osmo-sccp code, sccp_connections
was used in the NAT for tracking a sccp_connection. Rename it so it is
obvious that the struct belongs to the nat.
The rename was done with sed:
$ sed -i s,"struct sccp_connections","struct nat_sccp_connection",g \
include/openbsc/*.h src/osmo-bsc_nat/* tests/*/*
Prepend the international number with a '+' and then do the normal
re-writing on it. There are a couple of ways to handle this:
\+([0-9]), \+[0-9][0-9]([0-9]), \+49([0-9])
Add a test case for the international re-write based on an already
internationalized number.
We want to send a TRAP with the MGCP statistics from the NAT and
the connected BSC. The BSC endpoint can be either released because
of a DLCX from the MGCP CallAgent or the SCCP Connection release on
the A-link.
This is why we need to queue the statistics when the deleting the
endpoint on the BSC. The processing is continued once the response
arrives. This code assumes that the response of the DLCX will be sent
by the remote side. The current amount of outstanding responses can be
seen on the VTY. This assumption is based on the fact that the BSC has
already responded to the CRCX and maybe to the MDCX.
The MGCP RFC is bended to prefix the transaction identifier with "nat-"
to easily detect the response and hand it to the handler. This will
then parse the response and generate the TRAP. The current version is
v1. We assume that the transaction space is big enough and we will
not re-assign the transaction identifier too early.
In preparation for another kind of black-list allow the filter code
to decide how the connection should be rejected. Introduce a new struct
that will carry the reject causes for certain operations.
This file is created in ./configure so we shouldn't remove it with make.
Otherwise ./configure && make clean && make check fails with:
make[3]: *** No rule to make target `atconfig', needed by `check-local'.
Stop.
We parse the load_config, take the ptrdiff_t from start and load_config
and from the previous array as the alignment can differ on different ABIs.
This was found by Daniel when executing the tests on a 64 bit userspace.
MGCP is used over UDP and a response might be lost. The MGCP RFC
asks for keeping a list of responses and then using the previous
response to answer a duplicate request. I tried to conserve memory
and just wanted to remember the last transaction identifier and
result-code and re-generate the result from that. This made the
code look bad and this is why the entire response will now be stored.
It sadly increases the memory usage but can not be avoided at this
time.
Remove the msg->l3h pointer for the RQNT callback as strtok has
modified the content of it.
Calculate the expected packages and packet loss as of RFC 3550.
The values should be clamped but our packet loss counter is 32
bits and not 24 and we should clamp at other values but I am
waiting for some issues first before dealing with that.
I saw the old copy of the "Appendix J" code too late and I have
discovered some quirks and I am more familar with my implementation.
Most noticable 'w' only needs to be as big as the input arfcn but
requires the 'w' to be initialized. The power_of_2 implementation
differs as well (mine matches the output of wirehsark).
The f0 could be chosen in a better way but right now picking
the lower bound is the easiest. It is not clear if to use
modulo if the range is chosen in the middle. This can be improved
in the future. Right now I have no bit fiddling for range128, 256
and 1024 as I was running out of time.
Instead of building complex manual byte-wise parsers, we simply use two
strtok_r loops: one iterating over all the lines, the next one
iterating over the invididual space-separated elements in the first line.
The benefit is that we now accept \r, \n or \r\n, or any multiple of
them as line ending. This works around incompliant MGCP implementations
like that of Zynetix MSC.
Addition: mgcp_analyze_header returns 0 when all out parameters have
been set.
Signed-off-by: Holger Hans Peter Freyther <zecke@selfish.org>
Introduce number rewriting of SMS-SUBMIT. Introduce a new list,
move code around to help with finding a new number, somehow the
number encoding for TP-DA is borked, 03.40 references 04.11 but
the length appears to be strlen(number) without taken the type
field into account.
gsm_data_shared.h includes e1_input.h of libosmo-abis, add the
LIBOSMOABIS_CFLAGS to Makefile.am, remove AM_LDFLAGS at the same
time as we only build .a archives.
Match IMSI and destination address against a set of entries, if it
is matching the header will be modified and no sender report will be
requested. Change the test case to request the sender report and then
verify that this bit is reset to 0.
GSM 04.64 8.4.2 asks to ignore UI frames if the DLCI is not known,
or if the "(V(UR)- 32) <= N(U) < V(UR)". E.g. if we want to have
V(UR) == 511 and this frame is dropped, we would ignore N(U)'s
0 to 510. Calculate the delta.
The code is based on Jonathan Santos's "LLC UI window" fix but the
issue was discovered independly.
Move the regexp parsing code from the NAT to libcommon as it will
be used by the NAT and BSC code. This also adds the #include <regex.h>
include to gsm_data. This header should be split up.
This is a big patch that ports openBSC over libosmo-abis.
Sorry, the changes that are included here are all dependent
of libosmo-abis, splitting them into smaller pieces would
leave the repository in some intermediate state, which is
not desired.
The main changes are:
- The directory libabis/ has been removed as it now lives in
libosmo-abis.
- new configuration file format for nanoBTS and HSL femto, we
need to define the virtual e1_line and attach it to the OML
link.
- all the existing BTS drivers (nanoBTS, hsl femto, Nokia site,
BS11 and rbs2000) now use the new libosmo-abis framework.
- use r232 input driver available in libosmo-abis for bs11_config.
- use ipa_msg_recv instead of old ipaccess_read_msg function.
- delete definition of gsm_e1_subslot and input_signal_data.
These structures now lives in libosmo-abis.
Most of this patch are deletions of libabis/ which has been
moved to libosmo-abis.
This patch also modifies openBSC to use all the new definitions
available in libosmocore and libosmo-abis. In order to do that,
we have replaced the following:
- DINP, DMI, DMIB and DMUX by their respective DL* correspondences.
- SS_GLOBAL by SS_L_GLOBAL
- SS_INPUT by SS_L_INPUT
- S_GLOBAL_SHUTDOWN by S_L_GLOBAL_SHUTDOWN
- SS_INPUT by SS_L_INPUT
- S_INP_* by S_L_INP_* sub-signals
- E1INP_NODE by L_E1INP_NODE vty node
This patch has been tested with:
- one nanoBTS
- the HSL femto with the examples available under libosmo-abis
- BS11 with both dahdi and misdn drivers.
GCC 4.6.0 and LD.BFD 2.21 on ARM somehow fail to resolve
the dbi symbols when we have the library in front of the
static libraries, move them to the back.
Without this patch the tlv_def_patch symbol and the
gsm48_construct_ra.
GCC 4.6.0 and LD.BFD 2.21 on ARM somehow fail to resolve
the dbi symbols when we have the library in front of the
static libraries, move them to the back.
The idea that MCC and MNC is enough to classify a subscriber
turns out to be wrong. Certain operatos license a number range
of IMSIs to others. When we see a '^' in the MCC field we treat
it as a regexp. The code now turns the MCC/MNC into a regexp
for the IMSI. It is not using extended POSIX regexp to match
the behavior of the access list.
Use the libosmocore code to ignore certain signals by
default (e.g. SIGHUP, SIGPIPE) and use the new code to
create a default stderr logging target and initialize
it properly.
The current code tries to find _one_ bsc for a paging message
and then continues. The new code will try to find multiple BSCs
for each LAC. This is done in preparation of having two BSCs
handle the same LAC. This code right now is O(m*n) but it will
be worse once paging groups are landed.
The code to test the function was reduced to just test the lac
lookup code as the other part can not be tested in a standalone
setup anymore.
In 136f453dd2, I forgot to update
the new header path in the test/bsc-nat files (this happened to
me because I forgot to configured openBSC without --enable-nat).
libosmogsm is a new library that is distributed in the libosmocore.
Now, openbsc depends on it. This patch gets openbsc with this
change.
This patch also rewrites all include path to the new
osmocom/[gsm|core]
Signed-off-by: Pablo Neira Ayuso <pablo@gnumonks.org>
On a classic BSC we have 32 channels but one is reserved for
signalling. Make sure that we are not assigning 0x1f as we
assume that this is the signalling channel. This means that
from 32 possible voice channels we are only going to use 30
as we are already not using the 0x0.
Allocate the status for an endpoint dynamically. We will support
BSCs with different amount of multiplexes and need to have this
flexibility in the future. Add the proper null checks to the
current users of this code.
bsc_init.c was a big mess even only for two supported BTS models,
so before adding more BTS types, this needs a cleanup.
All the BTS specific code from bsc_init.c has now moved into
bts_{siemens_bs11,ipaccess_nanobts}.c
This has required that input_event() and nm_state_event() get both
converted to proper libosmocore signals instead of referencing external
symbols.
If one paging request is timing out the others will timeout soon
as well. With the current code the next timeout would expire the
next request in the queue. We will now stop all paging requests
and then issue a next paging request. So for both paging success
and failure we will now stop all the other requests.
This is mostly a workaround, one should count on how many
BTSes we are paging and wait for all failures before we remove
the item from the queue.
The reason for this is quite simple: We want to make sure anyone
running a customized version of OpenBSC to operate a network will
have to release all custom modifiations to the source code.
Extract the IMSI from the first message as well and safe it
in the connection structure. The problem is that we do not
have this structure at this point, so we will allocate the
imsi as child of the bsc_connection and then move/steal it.
Currently it is not is not easily possible to disable
everyone and then only allow certain SIMs. By changing
the order we can do:
access-list imsi-deny only-something ^[0-9]*$
access-list imsi-allow only-something ^123[0-9]*$
and still keep the usecase of only forbidding certain
SIMs on certain LACs. Adjust test case, test that the
other cases are still functional.
Remove a lot of code in favor of a new function that is freeing
the old string and copying the new one. I should have gotten the
context and the strings right.
Make it possible that one BSC is serving multiple
cells. Introduce a list of lacs, add functions to
manipulate the lists. The current test cases for
paging by lac continue to work.
Store the mapping from request to channel type in the GSM Network
struct as there is some policy involved with handling the request.
E.g. in a half rate network we don't want emergy calls to be getting
a TCH/F, or we want to have a different policy for early/late assignment
of phone calls. Update the table when creating the network and when
the neci is changed.
We are going to have more than one trunk, so all code hardcoding the
multiplex to zero must go. Avoid this kind of problem by saving the
MGCP endpoint number and comparing that.
common_vty.c was including bsc_nat.h which tried to
get the sccp/sccp_types.h which is not required to be
installed. Move all structs using/embedding SCCP structures
into the bsc_nat_sccp.h and include. This should fix
the compilation.
Add --enable-nat and --enable-osmo-bsc to build applications
requiring the Osmo SCCP library to be installed. We are not
using autodiscover as this is out of fashion.
Inside the access-list we have a list of entries that have
either one allow or one deny rule... we do not allow to remove
a single rule but one has to remove the whole list, in that case
talloc will handle cleaning all entries.
Right now the matching is O(n*m) as we traverse the list
(multiple times) and run the regexp multiple times. One
way to make it faster would be to concat all regexps into
one.
One can set one access-list to one BSC and one
access-list to one NAT. The matching of IMSIs
remains the same for now, also applying the
white/blacklist. Access lists can not be deleted
for now and no perf opt is done (e.g. one could
cache the result of the last lookup in the bsc
struct).
Right now it was not possible to just find a connection, by returning
the connection that is created we will have direct access to it. It
will be used by the local connection handling.
Now we are parsing a CM Service Request, Location Updating Request
and the Paging Response. For all other messages we claim to not
support it and force a refuse.
For now we have:
1.) bsc imsi deny to deny at the BSC level
2.) bsc imsi allow to allow a SIM at the BSC level
3.) nat imsi deny to deny at the global level
When setting a new MSC timeslot to a SCCP connection check if
any of the existing connections have this timeslot, if so we will
send a DLCX down the stream to make sure it is closed there, when
we will CRCX this new timeslot we will happily reallocate it.
When the SCCP connection goes away, or we get a DLCX from the
network, or the BSC is gone we will send a DLCX message down the
stream as well.
When we receive a CRCX from the network we will forward the CRCX
as usual and send a dummy MDCX after it.
For the DLCX and the dummy MDCX we send a custom MGCP message
that will not provoke an answer. Even if the downstream MGCP GW
will answer we will ignore it due the dummy transaction id that
is not used anywhere else.
This change should make sure that we close the dowstream endpoint
all the time, even when the DLCX arrives after the SCCP connection
is torndown.
When sending a MSG to the MSC try to find the to be used "src" reference
by comparing the reference on the BSC and the BSC connection. Only this
tuple needs to be unique.
Actually only when looking at the SRC REF we need to compare the BSC as the
dest reference should be unique but we are just making the check a bit stronger
to make it look symmetric.
We will reset the multiplex in a DLCX message and then
we can reset the multiplex as well...even if the MGCP
connection is staying open. or at least this is a theory.
The MSC likes to leave a connection open during CallControl
when hanging up early enough in the process.
Remove the code to parse port as we need to discover the
BTS behind the nat and most likely it will have a different
port than the one advertised by the BTS.
This reverts commit c6a1fe773d.
Fix the test to search for the original message instead
of the already patched one that should not find any items
anyway.
The remove is called on already patched connections so we
need to match it with the patch reference count.
Instead of checking the token for NULL we need to check if running
was set to null. Look at the data of the token and check if the line
was ending with a \r\n or \n and then when rewriting a line use that
line ending as well. Add a new test for that.
Holger Hans Peter Freyther
Alexander Huemer
Harald Welte